Decision Trees as a Method for Forecasting Seizure Precipitants and Identifying Their Influences on Seizure Outcome

Abstract

- Epilepsy is a complex disease that causes unpredictable seizures, which can lead to severe neurological impairments. Not knowing when a seizure will occur, many people with epilepsy often experience feelings such as anxiety, fear, and stress. In an effort to predict when seizures might occur, investigators have used data from patients’ electronic seizure diaries, as well as machine-learning methods, like decision trees. The objective of this work is to create patient-specific decision trees to 1) forecast seizure occurrence and identify seizure precipitants that influence seizure occurrences, and 2) determine seizure precipitants’ level of influence on seizure occurrences. Patients’ (n=64) seizure diaries were examined individually. Diaries contained data on how patients rated mood, predictive symptoms, stress, seizure occurrences, and seizure likelihood using a 5-point Likert scale. Diaries were recorded in the morning and in the evening, thereby evaluating seizures by half days. R Programming software was used for data analysis and decision tree development, and a confusion matrix was used for predictive accuracy. Results showed that precipitants’ influence on patient’s seizure outcome was greater in the morning than in the evening. Patients were also categorized in groups based on shared seizure precipitants. This work introduced non-invasive, personalized healthcare regimen for people with epilepsy.